keras model.summary() don't show all model layers

Question

I build a keras functionnal model and when I plot the summary, the first conv1d layer don't appear...

from tensorflow.keras.layers import Input, Dense, LSTM, Dropout, TimeDistributed, Conv1D, 
     MaxPooling1D, Flatten
from tensorflow.keras import Model, regularizers, initializers

tensor_input = Input(shape=(Xn.shape[1], Xn.shape[2]), name='main_inputs')

xy = TimeDistributed(Conv1D(filters= 10, kernel_size= 3,
                            activation=params['activationCNN1']))
xy = TimeDistributed(MaxPooling1D(pool_size= 2))

xy = TimeDistributed(Conv1D(filters=5, kernel_size= 2,
                        activation=params['activationCNN1']), name='Cnn1d-2')
xy = TimeDistributed(MaxPooling1D(pool_size= 2), name='MaxPool')
xy = TimeDistributed(Flatten(), name='Flatten')

xy = LSTM(params['unitsLstm1'],activation=params['activationLSTM1'],
          return_sequences=False, stateful=params['stateful'], 
          name='Hlayer1')(tensor_input)
xy = Dropout(rate = params['dropout1'])(xy)

xy = Dense(params['unitsDense1'], activation=params['activationDense1'],
           kernel_initializer= initializers.he_uniform(), name='Dense1')(xy)
xy = Dropout(rate = params['dropout2'])(xy)

out = Dense(autres_param['timestepsOut'], activation=params['activationDenseOutput'],
            kernel_initializer= initializers.he_uniform(), name='DenseOutput')(xy) 
model = Model(inputs=tensor_input, outputs=out)

model.compile(optimizer=optimizer, loss=params['loss'])
# summarize layers
print(model.summary())

What I get is: only the input, lstm, dropout and dense layers...

The trainning seem to work but all layer are active? How can I get the full summary??

Answer 1

You need to call each layer on the previous ouput tensor:

tensor_input = Input(shape=(Xn.shape[1], Xn.shape[2]), name='main_inputs')

xy = TimeDistributed(Conv1D(filters= 10, kernel_size= 3,
                            activation=params['activationCNN1']))(tensor_input)
xy = TimeDistributed(MaxPooling1D(pool_size= 2))(xy)

xy = TimeDistributed(Conv1D(filters=5, kernel_size= 2,
                        activation=params['activationCNN1']), name='Cnn1d-2')(xy)
xy = TimeDistributed(MaxPooling1D(pool_size= 2), name='MaxPool')(xy)
xy = TimeDistributed(Flatten(), name='Flatten')(xy)

xy = LSTM(params['unitsLstm1'],activation=params['activationLSTM1'],
          return_sequences=False, stateful=params['stateful'], 
          name='Hlayer1')(xy)

In general, the summary is good at showing all part of the model. If something is not shown in the summary, it's safe to assume that that means that it's not in the model.

Answer 2

In

xy = LSTM(params['unitsLstm1'],activation=params['activationLSTM1'],
          return_sequences=False, stateful=params['stateful'], 
          name='Hlayer1')(tensor_input)

you call the LSTM on tensor_input, so basically all this:

xy = TimeDistributed(Conv1D(filters= 10, kernel_size= 3,
                            activation=params['activationCNN1']))
xy = TimeDistributed(MaxPooling1D(pool_size= 2))

xy = TimeDistributed(Conv1D(filters=5, kernel_size= 2,
                        activation=params['activationCNN1']), name='Cnn1d-2')
xy = TimeDistributed(MaxPooling1D(pool_size= 2), name='MaxPool')
xy = TimeDistributed(Flatten(), name='Flatten')

is ignored.